RELIABILITY OF SOME ORE CHARACTERIZATION TESTS

*R. Chandramohan, G. S. Lane, B. Foggiatto, M. P. Bueno

Ausenco

144 Montague Road, South Brisbane

Australia 4101

(*Corresponding author: rajiv.chandramohan@ausenco.com)

RELIABILITY OF SOME ORE CHARACTERIZATION TESTS

ABSTRACT

Ore characterization is integral to any flowsheet development and equipment selection. There are

risks to the overall flowsheet development if sufficient feed material is not characterized adequately

representing the overall mine plan. The paper highlights some potential issues and opportunities to improve

some of the commercial characterisations test procedures when macro-scale fractures, such as those

apparent in RQD (Rock Quality Data); the response of friable ores with altered (argilic material) and

remanent unaltered and hard / competent rock; and as well as variable rock shapes that make up the feed to

a comminution circuits are considered.

KEYWORDS

Ore characterization, comminution, drop weight testing, RQD, friable ores

INTRODUCTION

Quantifying ore characterization is integral to any mine-project development. Over the years, a

number of ore breakage characterization tests have been developed specific for each comminution stage.

These specific tests were developed based on the understanding of machine dynamics and energy-size

relationship of the ore to be comminuted. However, there is nosingle ore characterization test that cover all

aspects of comminution (i.e. from blasting to fine grinding). The insight obtained between energy and

crushed product (quantity and size) was a leap in comminution understanding and model development,

(Hukki, 1961).

Lane et al. (2013), provided a comprehensive summary of various power-based comminution

models that are currently used in design and circuit optimisation studies; the authors highlighted that the

comminution power calculations are dependent on the consistency and reliability of the various

commercial ore characterization tests that aim to model each comminution stage.

A number of technical reviews have been completed on the small-scale comminution methods for

the application of flowsheet development and comminution modelling (Angove and Dunne, 1997), Mosher

and Bigg (2002), Bailey et al. (2009) and Morrell (2009)). These reviews highlighted important issues

pertaining to some commercial tests, especially the tests aimed at AG / SAG mill size selection, where the

results can be biased by the various test laboratories for the same ores (Bailey et al., 2009). There are risks

in the design and flowsheet, if ore characterization results and data of these tests are not well understood.

An example of this is the coarser ore strength measurement procedure determined using the drop weight

tester. The calculated Axb strength parameter can be skewed by 10% due to misinterpretation of the test

work data, hence leading to incorrect SAG specific energy calculations for competent rocks or Axb values

< 30 (Bailey et al., 2009). These variations in the calculated Axb can either be due to: the operation of the

drop weight tester; representative coarse particle selection; and or due to the calculation method used to

determine Axb from the t10 vs. Ecs data. Additionally, Bailey et al (2009), commented that the variability

of the feed size distribution may also have impact on the measured SAG specific energies.

Other highlighted issues pointed out by Bailey et al. (2009) pertain to the suitability of some tests

for competent ores. The authors commented that the SPI test is more suitable for less competent ores and

whereas the JK drop weight tests are better suited for more competent ores. Additionally, the SPI test may

be biased by sampling for multicomponent ore. This issue was recognized by Starkey et al. (2006) who

later developed and improved the SPI test with the SAG Design Test for competent ores.

This paper presents a review of some commercial tests and highlights potential issues when

interpreting comminution testwork data for the purpose of flowsheet development. The review specifically

focused on the issues of:

Representative sample selection – particle shape

Influence of macro scale rock fracture on competency and SAG feed size

The response of soft friable ores in comminution tests – Bond grindability test

SUMMARY OF ORE CHARACTERIZATION TEST WORK METHODS AND POWER BASED

MODELS

When Bond first developed his comminution law, crushers, rod mills and ball mills were common

equipment in the flowsheet. These tests were developed for the basis of scale-up of staged crush, rod and

ball mill circuits (Bond, 1952). With the introduction of AG/SAG milling, HGPR and fine grinding in the

flowsheet, specific comminution tests followed to suit these devices. Table 1 shows a summary-list of

commercial comminution tests used in various comminution models. The applicability of these tests for the

purpose of comminution circuit development in various comminution models was summarised by Lane et

al. (2013).

Table 1 – Summary of small-scale comminution tests (modified after Mosher and Bigg (2002))

Test Name / Reference Top Size

(mm)

Sample

Requirement

(kg)

Type / Steady-state Comminution Device Test

Aim

Advanced Media Competency Test (Siddall et al., 1996) 165 300 Batch / No AG / SAG Mills

UCS (Ulusay and Hudson, 2007) NX, HQ, PQ cores with

1:~2.5

diameter / core length

ratio

Min 10 cores Core / No Crushers

Point Load (Ulusay and Hudson, 2007), (ASTM, 1985, Bearman, 1999)

40 – 30 20 Single particle / No Crushers

Bond Crusher Test (Bond, 1952) 75 10 Single particle / No Crushers

JK Drop Weight Test (Napier-Munn et al., 2005) 64 75 Single particle / No Crushers, AG / SAG mills

SMCC Test (Morrell, 2004) 31.5 5 Single particle / No AG / SAG mills

SAG Design Test (Starkey et al., 2006) F80 = 152 10 Batch / No AG / SAG mills

MacPherson Test (MacPherson and Turner, 1978) 32 100 Continuous / Yes AG / SAG mills SAG Power Index (SPI) (Starkey and Dobby, 1996) 19 5 Batch / No AG / SAG mills

Lab HPGR Test (Daniel, 2007) 12.5 25 Continuous / No HGPR

Bond Rod Test (Bond, 1952) 13 10 Locked-cycle / Yes Rod Mills, AG / SAG mills Bond Ball Test (Bond, 1952) 3.5 10 Locked-cycle / Yes Ball Mills

Stirred milling Test1 ~1.5 10 – 20 Continuous / No Stirred mills

ISAmill Test2 ~1.5 10 – 20 Continuous / No ISA mills

Tests such as the Unconfined Strength (UCS) and Point Load Index (PLI) measure the bulk and

tensile strength of the ore. The degree of broken product is not quantified in these tests. Whereas, tests such

as the JK drop weight and the SMCC tests quantify the strength of ore using product size-energy

relationship. The Advanced Media Competency, SPI, SAG design and MacPherson tests aim to quantify

the specific energy required for AG / SAG milling by using lab-scale mills. For rod and ball mill tests, the

standard Bond rod and ball mill tests are followed. The HGPR and fine grinding tests use customized lab-

scale units to quantify the machine performance under variable operating conditions.

1 Stirred milling tests are machine specific

2 ISA milling tests are machine specific

As reported by Lane et al. (2013), some comminution models use a combination of JKSimMet

(JKTech, 2007) and in-house empirical models developed from plant data to predict the power and

throughput requirements of various comminution circuits. The AusGrind calculation is one such power-

based model used extensively by Ausenco, (Lane et al., 2013). The calculation relies on empirical

relationships obtained from +20 years of pilot and actual plant data to predict the comminution energies

and throughputs for various circuits.

The use of AusGrind in flowsheet development follows an initial assessment and analysis of the

test work data, review of project scale and geology and benchmarking other similar projects and risk

assessments (Error! Reference source not found.). The aim of test work is to address the key risks and

opportunities as a function of project scale and geology. The most common parameters and their associated

test methods used by AusGrind for throughput and comminution power calculations are:

DWi from SMCC tests

Axb from JK drop weight tests

CWI from Bond Crusher tests

RWI and BWI from the Bond rod and ball mill tests

Other testwork data such as the Advanced Media Competency tests, SPI and SAG Design test

empirical relationships benchmarked against DWi or Axb results are used.

Figure 1 – Ausenco’s comminution design methodology (modified after Lane et al. (2013))

Misinterpretation of the above mentioned ore breakage characteristics can lead to inaccurate

comminution energy calculations and throughput predictions. Circuit energy and throughput calculations

should be benchmarked against actual plant performance and based on breakage characterization tests that

are reliable. Also, the impact of ore variability on calculated comminution power need to be well

understood. Some of the factors that impact on the comminution and throughput calculations can be

attributed to:

Sample representation

o Extreme sample variability with a mix of extreme hard and soft ores

Inadequate coarse ore particle selection

o The interpretation of rock fracture, fracture in-fill type and macro-size competency,

(Barratt, 2009).

o High particle shape variability

In order to minimise the risk in flowsheet development, pilot plant trials are required to provide

deeper insight into the comminution performance prior to defining the process design criteria. These can be

used to assess ore variability but the use in this context is limited due to the high cost associated to pilot

plant sampling and operation. These trials can highlight some of the potential sensitivities and risks that are

not captured in the ore characterization tests. A typical (SABC) small-scale pilot plant trial may require a

20-tonne representative sample to achieve steady-state operation, which can be difficult to obtain especially

if trials are conducted to evaluate the circuit performance for variable ore feed. Therefore, the reliability of

small-scale comminution tests require careful interpretation to (a) select ore feeds to be tested and (b)

reduce potential project risks.

REPRESENTATIVE SAMPLE SELECTION – PARTICLE SHAPE

The comminution strength parameters such as the Axb (from the JK drop weigh test method) or

DWi (from the SMC test method) rely on non-linear regression analysis to fit empirical models to the t10

vs. Ecs data. Chandramohan (2010), (2011) and (2013) emphasized that the interpretation of the JK drop

weight data can be misleading when biased sample shapes are selected to calculate the Axb parameter.

Figure 2 shows sample test work data for non-flaky and flaky particles orientated horizontally and

vertically on the drop weight anvil.

Sample shapes were selected according to the techniques developed by Nakajima et al. (1978).

Table 2 shows the calculated Axb values for varying particle shapes for the same ore. According to the

standard t10 calculation method (Narayanan and Whiten, 1988), the t10 value is considered to be the 1/10th

passing of original parent particle mean dimension. For simplicity, the t10 value is calculated from the mean

retained mesh screens, which is based on the assumption that the selected particle is ‘blocky’ or non-flaky

in shape. Therefore, when flaky rock shapes orientated horizontally were corrected for the actual particle

thickness, the corrected Axb values are reduced, indicating a more competent shape than the non-flaky

samples. No clear protocols are in place to minimize the shape biases when selecting particles according to

the standard JK drop weight procedure. Stark et al. (2008) noted the importance of operator training and

monitoring when conducting the JK drop weight tests, where biased sample selection could lead to

uncertainty in the final result.

Table 2 – Calculated Axb values based on particle shape and orientation

Shape Axb

Non-flake 63.7

Flake vertical 59.0 Flake horizontal 91.7

Flake horizontal (corrected) 53.4

Figure 2 – Data for samples tested based on shape (Chandramohan (2013))

To minimise the variances in the JK drop weight result, Chandramohan et al. (2013) proposed a

mechanistic model to predict the Axb value for any rock shape. The mechanistic model, shown in equation

(1) and (2), uses known-controlled shapes, in this case biased blocky or non-flaky shapes, to calibrate the

shape factor for any rock shapes. In equation (2), the is the UCS strength value; is the internal

friction angle or rock (typically 31 degrees for most rocks); and are related to the internal friction and

angle and dynamic friction; and Y is the particle thickness resting on the anvil.

n

nNonflakeFlake

akeFModelNonFl

FModelFlakeAxbAxb

Equation (1)

R

UCS

n

Y

F

1

sin1

sin1

Equation (2)

Chandramohan (2013) recommended using the SMCC test protocols to deliberately bias the non-

flaky samples for model calibration. The particle-selection criteria in the SMCC test is tightly controlled by

the mean density. Therefore, the sample-shape biases are minimized (Morrell, 2004). Using the knowledge

of variable competencies for variable rock shapes, Chandramohan et al. (2011) presented potential

opportunities to take advantage of rock shapes for the benefit of AG/SAG optimisation. This work showed

the potential application to control grind and throughput by maintaining optimum flaky and non-flaky

shape-ratio in the SAG feed. Figure 3 illustrates the steps to remove the bias in the calculated Axb values

based on particle shape. The flaky shapes are removed from the sample using slotted screens that are 42%

of the passing screen-mesh size. The production of flaky-shapes is dependent on crusher operation,

(Bengtsson and Evertsson (2006), (2009)). Therefore, for flowsheet development, it is useful to quantify

the amount flaky product produced from crushers. For higher proportion of flaky product, the overall Axb

value should be calibrated to match the likely SAG feed competency based on the proportion of particle

shape when sizing AG / SAG milling equipment.

Figure 3 – Decision tree for selecting particle shape and quantifying Axb strength, (Chandramohan, 2013)

THE INFLUENCE OF ROCK FRACTURE ON COMPETENCY AND SAG FEED SIZE

The RQD measurements from drill core logging are an indicator of the ore profile rock quality

down a drill hole. Error! Reference source not found. shows an example of the procedure for measuring

RQD in cores. High RQD values indicate broken core and low RQD values indicate unbroken core.

Figure 4 – Procedure for measuring and calculating RQD (after Deere and Deere (1988))

Examples of lower and higher RQD values cores are shown in Error! Reference source not

found.. In this example, the lower RQD value is 53 % and the higher value is 87%. As shown in the

figure, the sample with lower RQD comprises of significantly fractured cores and fines. The sample with

higher RQD value contains less fractured material. The estimated RQD for the cores are based on visual

inspection; therefore the measured values are subjective and are susceptible to errors in the estimating

method.

Figure 5 – Examples of core RQD

In regards to hardness measurements, to-date there has been no conclusive evidence to suggest

that RQD has any significant impact on measured coarse-ore competency parameters such as the JK drop

weight’s Axb and SMCC’s DWi values. Wirfiyatal and McCaffery (2011) developed a throughput-

relationship model based on geotechnical measurements such as PLI, RQD, Rock Mass Rating (RMR) and

comminution parameters such as JK drop weight and Bond tests for Batu Hijau operation. In their work,

Wirfiyatal and McCaffery (2011) commented on the considerable limitation of the throughput relationship

prediction when using RQD benchmarked against PLI. The main issue with using PLI as a proxy for

hardness (fracture toughness) for SAG throughput calculation was the limitation in the size fraction tested.

The PLI test uses 35 – 65 mm size rocks, which account for approximately 15% of SAG mill energy

consumption, (Wirfiyatal and McCaffery, 2011). In order to determine the overall SAG energy

consumption, representable SAG feed size distribution is required. The single particle drop weight tests

used to calculate DWi and Axb parameters, incorporate wider representable SAG feed distribution. It was

suggested by Wirfiyatal and McCaffery (2011) for the Batu Hijau geology, that the quartz-vein density

drives the mineralogy and hardness. Therefore, for homogenous mineralisation, larger particles have

similar hardness to smaller particles. As the quartz density decreases, so does the mineralisation, hence

reducing the similarity in hardness between large and smaller rock particles. Since the drill cores are either

crushed or cut into test pieces prior to competency tests, it is hypothesized that mineralogy and density of

fractures between coarse and fine particles heavily influences the measured Axb or DWi values, rather than

theRQD values of the cores (Barratt, 2009, Wirfiyatal and McCaffery, 2011). Therefore, the SAG

throughput and power predictions are mainly influenced by the SAG feed due to the inherent macro scale

rock fractures and densities.

IMPACT OF PARTICLE SELECTION ON COMPETENCY

Figure 6 and Figure 7 illustrate two examples of drop weight data for low and high RQD samples

and the analysis of the data shown in Table 3 and

Table 4. Fitted curves of lower bound, upper bound and best-fit curves are shown using the

Narayanan and Whiten t10 equation, (Narayanan and Whiten, 1988). The t10 value is not constant for

varying particle sizes as shown by the variability in the specific energy graphs (t10 vs. Ecs), especially for

high impact specific energy tests (above 1 kWh/t). The t10 values are higher for the coarser particles than

for the finer particles. For the high scatter sample (low RQD), the distribution of the t10 data for high-

energy impact is significant (standard deviations shown in Table 3). These observation were also noted by

Morrell in his earlier works (2004) and (2009). For the low scatter sample (high RQD), in which the t10

distribution for high-energy impact is low (standard deviations shown in

Table 4).

Figure 6 – Low RQD sample (high scatter) drop weight data, Best fit Axb = 36.7

Figure 7 – High RQD sample (low scatter) drop weight data, Best fit Axb = 29.2

In this example, Axb values are calculated using Narayanan and Whiten equation and shown in the analyses data, (Table 3

and

Table 4). The calculated lower and upper bound Axb show significant variability when compared

with the estimated best-fit values. The upper bound Axb values, estimated from the highest t10 values per

tested specific energies are higher, indicating a less competent ore. The lower bound Axb values estimated

from the lowest t10 values per tested specific energies are lower, indicating a more competent ore. The best-

fit Axb values lie between lower and upper bound range. To determine extent of t10 scatter on the

calculated Axb values, 75th

and 25th

percentile values of the t10 data for the three energies are used to

calculate the Axb values.

Table 3 – High scatter (low RQD) breakage data

Size fraction

(mm)

Ecs

(kWh/t)

Best fit

t10 (%)

Upper bound t10

(%)

Lower bound

t10 (%)

t10 75th

Percentile (%)

t10 25th

Percentile (%)

16 x 13.2 2.5 40.4

56.7 40.4 53.3 45.1 22.4 x 19 2.5 49.9

31.5 x 26.5 2.5 56.7

Average 49.0

SD 8.2

16 x 13.2 1 16.9 39.0 16.9 34.1 23.4

22.4 x 19 1 25.6

31.5 x 26.5 1 32.4

45 x 37. 5 1 39.0

Average 28.5

SD 9.5

16 x 13.2 0.25 4.1

9.4 4.1 9.4 7.9

22.4 x 19 0.25 7.9

31.5 x 26.5 0.25 9.4

45 x 37. 5 0.25 8.6

63 x 53 0.25 9.4

Average 7.9

SD 2.2

Axb 36.7 54.2 17.3 46.8 29.6

Table 4 – Low scatter (high RQD) breakage data

Size fraction

(mm)

Ecs

(kWh/t)

Best fit

t10 (%)

Upper bound

t10 (%)

Lower bound

t10 (%)

t10 75th

Percentile (%)

t10 25th

Percentile (%)

16 x 13.2 2.5 37.9

46.3 37.9 46.2 42.0 22.4 x 19 2.5 46.1

31.5 x 26.5 2.5 46.3

Average 43.4

SD 4.8

16 x 13.2 1 18.4

28.8 18.4 26.1 19.8 22.4 x 19 1 20.2

31.5 x 26.5 1 25.2

45 x 37. 5 1 28.8

Average 23.2

SD 4.7

16 x 13.2 0.25 5.4

8.6 5.4 8.3 6.1

22.4 x 19 0.25 8.0

31.5 x 26.5 0.25 6.1

45 x 37. 5 0.25 8.3

63 x 53 0.25 8.6

Average 7.3

SD 1.4

Axb 29.2 39.8 21.4 34.3 22.8

Figure 8 shows the calculated Axb values from Table 3 and Table 4. Boxes and whisker plots are

used to show the variability Axb data. The whiskers show the calculated lower and upper bound values.

The boxes show the calculated 75th

and 25th

percentile values. The scatter in the calculated Axb value for

both samples is significant. The Axb ranges from 17.3 - 54.2 for the high scatter (low RQD) sample and

ranges from 21.4 – 39.8 for the low scatter (high RQD) sample. The 75th

and 25th

percentile Axb values

show wider distributed hardness range for the high scatter (low RQD) sample, indicating similar proportion

of soft and hard particles were selected for the JK drop weight tester. Whereas the 75th

and 25th

percentile

Axb values for the low scatter (high RQD) sample, show a skewed hardness distribution, indicating a

higher proportion of competent particles were selected for the JK drop weight tester.

Figure 8 – Analysis of the t10 data for low and high scatter samples

Figure 9 and Figure 10 Error! Reference source not found. show more examples of the

variability in the measured t10 vs. particle size and calculated t10 vs. Ecs curves for eight JK drop weight

samples. The scatter in the measured t10 is significant for varying tested specific energies, hence having an

impact in the measured Axb values (shown in Figure 11).

Samples 1 to 4 show variable t10 values at different particle sizes for energies tested greater than 1

kWh/t. Hence, these differences in the measured t10 values have an impact on the calculated t10 vs. Ecs

curves which are then used to determine the Axb values of the ore sample. For samples 5 to 8, the

measured t10 variability is mainly between the 0.25 to 2.5 kWh/t tested energy range. For these samples, the

particles tested at the 1 kWh/t energy show considerable t10 variability, hence having an impact on the

calculated Axb values.

In Figure 11 for the calculated Axb value distribution, samples 1, 6, and 8 have wider percentile

distributions and the rest with lower percentile distribution between the 75th

and 25th

. For samples that have

wider percentile distribution, it is probable that the samples comprised of even distribution of hard and soft

particles, which was observed earlier for the high scatter sample (low RQD) shown in Figure 6. For

samples with narrower and skewed percentile distributions about the best-fit Axb value, it is probable that

biased particle selections were used for the tests (samples 2, 5, and 7).

Figure 9 – Examples of best fit, upper and lower bound Axb data for variable samples (samples 1 to 4)

Figure 10 – Examples of best fit, upper and lower bound Axb data for variable samples (samples 5 to 8)

Figure 11 – Calculated Axb values for samples 1 to 8

As shown in the analysis above, representable and controlled particle selection is critical to the

analysis and calculation of Axb values. The distinct variability shown for the measured t10 values at each

tested energy for varying particle sizes have an overall impact on the calculated Axb range for each

sample. Theses variabilities in the calculated Axb values for each tested samples can have detrimental

impact on the overall SAG throughput and power calculations.

According to Burger et al. (2006), Barratt (2009) and Wirfiyatal and McCaffery (2011) fracture

densities in the sample have a significant impact on the measured hardness values. The coarser ore have

higher fracture densities, and therefore have higher Axb values indicating a less competent ore, and finer

ore particles comprising of lower fracture densities are considered to be more competent. Therefore

controlled particle selection protocols are required. Controlled particle selection method based on ore SG,

such as used in the SMCC test, will be useful to reduce the extreme particle selection biases (currently the

JK drop weight procedure do not use ore SG as a proxy for sample selection).

IMPACT OF RQD ON SAG FEED SIZE

Bailey et al. (2009) used Morrell’s equation (3), to determine the SAG feed (F80) or primary

crusher product based on ore competency and crusher closed side setting.

Equation (3)

Once the F80 of the comminution circuit is known, a suitable combination of equipment is

selected to achieve the desired design throughput. Bailey et al. (2009) commented on cases where the

predicted F80 from the above equation did not match crusher P80 for the desired CSS. In most of the

observed cases, the SAG mill feed was finer than expected and actual F80s were significantly lower than

the predicted ones. This mismatch in the predicted vs. actual SAG mill feed F80 can be either due to:

unaccounted inherent macro-scale rock fracture, or

reduced size distribution caused by higher intensity blasting or fragmentation during mining - for

the purpose of this paper, the influence of intensive blasting on ore competency is not

investigated.

The DWi is a measure of hardness based on ore SG and drop weight test data (Bailey et al., 2009)

It is the main factor in equation (3) that has an effect on the overall SAG F80 prediction. The DWi and/or

Axb values3 are important for AG/SAG mill power and throughput prediction. Therefore, the reliability of

the energy-based models is heavily dependent on representative competency measurements.

Database on RQD vs. SAG feed F80 was used to develop relationships to modify equation (3),

(Burger et al., 2006, Wirfiyatal and McCaffery, 2011, Morrell, 2014). Therefore equation (4) can be re-

written as for two conditions:

( ( ) ) ( ) If RQD > 25%

Equation (4) ( ( )

) ( ) If RQD < 25%

The best fit Axb values calculated in Table 3 and

Table 4 were used to estimate the impact of RQD on SAG mill feed F80 using equation (4), (Table 5 and

Table 6). For calculation purposes, a crusher CSS of 110 mm was used. As shown, the RQD reduces the

calculated SAG Feed F80 value but does not change the measured ore competency.

Table 5 – Estimating SAG F80 for Lower RQD data with best fit Axb data

Value Unaltered data (100% RQD) 53% RQD

Axb 36.7 36.7

DWi (kWh/t) 7.4 7.4

SAG F80 (mm) 89 65

Table 6 – Estimating SAG F80 for Higher RQD data with best fit Axb data

Value Unaltered data (100% RQD) 87% RQD

Axb 29.2 29.2

DWi (kWh/t) 9.2 9.2

SAG F80 (mm) 104 97

Figure 12 shows the decision tree for applying the techniques presented in this section. The RQD

factor is only applied to core samples. For core samples with varying RQD, the SMCC test recommended

to minimise sample selection variability. For crushed ore, either JK drop weight test or SMCC is preferred.

To estimate the SAG feed F80 for the core samples, equation (4) is recommended.

3 DWI is roughly proportional to the inverse of the Axb value for a given sample

Figure 12 – Decision tree for determining DWi and SAG mill feed F80 for throughput calculations of hard ores

THE RESPONSE OF SOFT FRIABLE ORES IN COMMINUTION TESTS

For friable ores, the size distribution is usually skewed to include high amounts of fines particles

in the feed. For competent and blocky ores, the size distribution is usually skewed with fewer fines.

Typically friable ore types have low competency in the coarse size fractions (Axb > 80). Low competency

ores are usually expected to have high throughput through the SAG mill and are circuits are constrained by

the secondary and tertiary grinding stages. Therefore, the risk in process design for the treatment of friable

ore is greater in the secondary or tertiary grinding stages.

The standard Bond ball mill work index test is conducted under dry milling conditions. For ores

with Bond work indexes greater than 12 kWh/t, dry locked cycle tests are suitable as they have low

difference in hardness between wet and dry milling. However, for friable ores that present high proportions

of clays, dry locked cycle tests may results inaccurate work index values and, therefore, either

modifications to dry test or wet milling is recommended. Man (2002) provided a detailed review on the

reasons of why the Bond grindability test is conducted the way it is. Man described the reasons for

conducting dry milling tests were mainly due to logistics of the test which may otherwise require cyclones

or other wet classifiers and pumps. Furthermore, Man (2002) noted that the effects of slurry viscosity may

bias the Bond grindability results. Dry milling tests are simpler and easier to carry out than wet milling.

However, in recent test work managed by Ausenco of soft ore project, the grindability result varied as

much as 20% between dry and wet milling; where the wet milling work index result was significantly

lower than that for dry milling and consistent with continuous milling pilot tests that were carried out.

To prevent “over-grinding”, rod milling is more suited than ball milling for friable ores that

contain high amounts of clays. Figure 13 shows an example of the product response in the standard Bond

rod and ball mill grindability tests. The ball mill tends to over-grind feed with a potential increase in slimes

production. A slimes increase of 5% in the -10 m fraction is estimated for ball milling compared to rod

milling, (Figure 13). Excess slimes or fines in the product can be detrimental in the flotation response; thus

decreasing the overall recovery. However, rod milling capacities are limited by the size of the mill. For

high throughput, lower operating costs, ball milling is preferred over rod milling.

Figure 13 – Example of Bond rod and Bond ball mill grindability test product for the same ore and feed size distribution

For ores that have excessive fines, a modified Bond grindability test is required. This is because,

the fine in the feed may bias the Bond milling work index. Therefore a pre-screened feed, shown in Figure

14, is suitable for the modified Bond grindability test. Key aspects of the tests are:

closing screens 1 and 2 have the same aperture.

ratio of fresh split is quantified between pre-screened feed to the locked cycle circuit and pre-

screened product.

test is conducted on the pre-screened feed targeting a circulating load of 250%.

o locked cycle tests have similar procedures as the Bond ball or rod mill tests for wet and

dry milling.

o for wet milling, a solids density of 70 % w/w is recommended. However, this can change

based on slurry viscosity impact on milling performance.

steady-state specific power of the locked cycle circuit is determined after a number of cycles

total specific power of the circuit based on fresh feed is then calculated from the ratio of pre-

screened feed and product.

the circuit Bond milling work index is calculated from the fresh feed F80 and the combined

product P80.

As shown in the method above, the calculated the Bond milling work index was based on fresh feed;

therefore a reverse closed circuit classifier is required for the circuit design.

Figure 14 – Locked cycle procedure for fresh feed with excess fines

As pointed-out by Man (2002), slurry viscosity has a significant impact the Bond work index

grindability value. Shi and Napier-Munn (1996), (1999), (2002) conducted extensive work on the impact of

slurry viscosity on ball milling performance. They developed a criterion for milling performance based on

the quantity of -38 m in the feed and product of the secondary and tertiary grinding circuits, shown in

equation (5).

(

) Equation (5)

Where:

is the percent passing -38 m in the mill discharge

is the percent passing -38 m in the mill feed

According Shi and Napier-Munn, the aim of the grinding index is to capture the inefficiencies of

the grinding and classification circuit. A high grinding index indicates higher events of particle-breakage

occurring inside the mil and a low grinding index indicates less efficient breakage occurring inside the mill

either due to inefficient cyclone classification (i.e. high circulation load) or due to high slurry viscosity.

Therefore, the grinding index can be used as a performance indicator to select correct cyclone spare parts

dimension (i.e. spigot and vortex finder) and operating pressures.

Figure 15 shows the logical setups required for selecting the type of test for friable ores. The aim

of the decision tree is to quantify and select the correct test procedure based the proportion fines in the

crushed product which may have an impact on the Bond milling work index and slurry viscosity.

Figure 15 – Decision tree for conducting Bond milling test on soft ores

CONCLUSION

The paper presented a review of some common commercial test work for coarse and friable ores.

The reliability of these tests and the understanding of the test outcome are pertinent for flowsheet

development. Tests such as the JK drop weight, SMCC and the Bond ball / rod grindability tests were

developed based on the necessity of comminution equipment selection. Therefore careful consideration and

understanding of the test work data is an important step in the design process. The main conclusions of

analysis of comminution tests sensitivities influence on flowsheet development are:

In the drop weight test procedure, it is assumed that sample selection for single particle tests is

random. However, a biased sample selection by shape can significantly change overall strength

index. As crusher product is considered to be representative of typical a SAG mill feed during

flowsheet development, particle selection for drop weight tests should reflect the variability in the

feed. If the crusher produces increased quantity of flaky ore, then the drop weight strength index

should be calibrated to reflect the proportion flaky and non-flaky shapes in the SAG mill feed.

Analysis of the t10 data showed significant scatter when uncontrolled sampling procedure is

followed when selecting particles for the drop weight tests. Coarser particles had higher Axb

values indicating less competent materials. Finer particles had lower Axb values, indicating more

competent samples. Commentary by Burger et al. (2006), Barratt (2009) and (Wirfiyatal and

McCaffery (2011), Burger et al., 2006) indicated that fracture density controls the hardness

distribution between coarse and fine particles. In terms of core RQD and its impact on ore

competency, particle selection is main driver for hardness variability. High and low t10 data for

each tested energy calculates the maximum and minimum range of Axb values. The 75th

and 25th

percentile values indicate the distribution of the data about the best fit value based on particle

selection. A wider distribution indicates an uncontrolled distribution of particle selection. To

minimise the uncontrolled bias in the sample selection for cores, it is highly recommended that the

SMCC test procedures and protocols should be followed to control the particle selection using ore

SG as a proxy. However, as the case with most commercial tests, obtaining representable sample

selection to determine overall ore competency will be a challenge, if the cost of the test is a

decision factor during the PFS.

Morrell’s (2009), SAG feed F80 calculation was modified to include variable RQD values. For

RQD values greater than 25%, a factored power-function relationship is used to calibrate the SAG

feed F80. For RQD values less than 25%, a constant factored multiplier is used. The conditional

RQD calibration of the SAG feed F80 was based on Ausenco’s RQD vs. SAG feed F80 database,

(Ausenco, 2014).

For friable ores, which contain high fines in the feed, careful operational consideration of the

secondary and tertiary grinding equipment is required. Therefore the Bond work index tests (for

ball and rod mill selection) should take into consideration of the high fines in the feed. A modified

Bond grindability test is required fines in the fresh feed to be removed prior to the test. For ores

with high amount of fines and Bond work indices less than 12 kWh/t, wet milling is

recommended. From authors’ experience, for these ores difference in the Bond work index of up

to 20 % between dry and wet milling has was observed.

The use of (Shi and Napier-Munn (2002)) Grinding Index is useful to optimise the target slurry

density of the ball milling circuit. They showed that the viscosity of the slurry has an impact on

grinding performance. Therefore, the Grinding Index could be used as a proxy for cyclone /

classification operating parameters.

Flowsheet development of comminution circuits must rely on representative testwork and good

understanding of the test work data. Risk mitigation plans to improve test reliability were proposed for both

hard and friable ores.

Three decision tree diagrams were developed as part of the procedure for comminution testing.

These logic-diagrams are recommended to be used as decision guidance when dealing with peculiar ores

such as the ones described in this paper.

ACKNOWLEDGEMENTS

The authors would like to acknowledge Ausenco for supporting the publication of this paper at the

SAG conference in Vancouver.

REFERENCES

ANGOVE, J. E. & DUNNE, R. C. A Review of Standard Physical Ore Property Determination. World

Gold Conference, 1997 Singapore. 139-144.

ASTM 1985. Suggested Methods for Determining Point Load Strength ASTM D5731-08. International

Society for Rock Mechanics Commission on Testing Methods, 22, 51 - 60.

AUSENCO 2014. Internal Document - RQD database vs. SAG mill feed F80. In: AUSENCO (ed.).

BAILEY, C., LANE, G., MORRELL, S. & STAPLES, P. 2009. What can go wrong wrong in

comminution circuit design? Tenth Mill Operators Conference. Adelaide, South Australia:

AUSIMM.

BARRATT, D. J. 2009. Technical MEMO - Crushing Work Index Comparison Between Labs. In: DJB

CONSULTANTS, I. (ed.).

BEARMAN, R. A. 1999. The use of the point load test for the rapid estimation of Mode I fracture

toughness. International Journal of Rock Mechanics and Mining Sciences, 36, 257-263.

BENGTSSON, M. & EVERTSSON, C. M. 2006. An empirical model for predicting flakiness in cone

crushing. International Journal of Mineral Processing, 79, 49-60.

BENGTSSON, M., SVEDENSTEN, P. & EVERTSSON, C. M. 2009. Improving yield and shape in a

crushing plant. Minerals Engineering, 22, 618-624.

BOND, F. C. 1952. The third theory of comminution. Transcript AIMME, 193, 484-94.

BURGER, B., MCCAFFERY, K., JANKOVIC, A., VALERY, W., MCGAFFIN, I. & LA ROSA, D. 2006.

Batu Hijau Model for Throughput Forecast, Mining and Milling Optimisation. In: KAWATRA, S.

K. (ed.) Advances in Comminution. Society for Mining, Metallurgy and Exploration.

CHANDRAMOHAN, R. 2013. Effect of Rock Shapes in Comminution. PhD, University of Queensland.

CHANDRAMOHAN, R., HOLTHAM, P. N. & POWELL, M. S. 2013. Development of a mechanistic

breakage model to predict the strength of varying rock shapes. European Symposium in

Comminution and Classification. Germany.

CHANDRAMOHAN, R., POWELL, M. S. & HOLTHAM, P. N. The infulence of paticle shape in rock

fracture. International Mineral Processing Conference 2010, 2010 Brisbane. AUSIMM.

CHANDRAMOHAN, R., POWELL, M. S., HOLTHAM, P. N., LANE, G. & DANIEL, M. J. 2011. The

Effect of Blends of Rock Shapes in AG / SAG Mills and Comminution Circuits. Semi-Autogenous

Grinding Conference 2011 - Fifth International Conference on Autogenous & Semiautogeneous

Grinding Technology. Vancouver, Canada: CIM.

DANIEL, M. J. 2007. Energy efficient mineral liberation using HPGR technology (PhD). PhD, University

of Queensland.

DEERE, D. U. & DEERE, D. W. 1988. The Rock Quality Designation (RQD) INdex in Practice,

Philadelphia, ASTM ST 984.

HUKKI, R. T. 1961. Proposal for a Solomonic settlement between the theories of von Rittinger, Kick and

Bond. Trans SME / AIME, 220, 403 - 408.

JKTECH 2007. JK SimMet users manual - version 5.

LANE, G. S., FOGGIATTO, B. & BUENO, M. P. Power-based Comminution calculations using

AusGrind. In: PROCEMIN, ed. Procemin 2013, 2013 Chile.

MACPHERSON, A. R. & TURNER, R. R. Autogenous grinding from test work to purchase of a

commercial unit. In: MULAR, A. L. & BHAPPU, R. B., eds. Mineral Processing Plant Design

1978 New York. AIME, 279 - 305.

MAN, Y. T. 2002. Why is the Bond Ball Mill Grindability Test done the way it is done? The European

Journal of Mineral Processing and Environmental Protection 2, 34 - 39.

MORRELL, S. 2004. Predicting the specific energy of autogenous and semi-autogenous mills from small

diameter drill core samples. Minerals Engineering, 17, 447-451.

MORRELL, S. 2009. Predicting the overall specific energy requirement of crushing, high pressure

grinding roll and tumbling mill circuits. Minerals Engineering, 22, 544-549.

MORRELL, S. 2014. RE: RQD vs. SAG Feed F80.

MOSHER, J. B. & BIGG, A. C. T. Bench Scale testing and pilot plant tests for comminution circuit design.

In: SME, ed. Mineral Processing Plant Design, Practice and Control, 2002.

NAKAJIMA, Y., WHITEN, W. & WHITE, M. 1978. Method for measurement of particle shape

distribution by sieves. Mineral Processing Extractive Metallurgy, 87, 194 - 203.

NAPIER-MUNN, T. J., MORRELL, S., MORRISON, R. D. & KOJOVIC, T. 2005. Mineral Comminution

Circuits - Their operations and optimisation, University of Queensland.

NARAYANAN, S. S. & WHITEN, W. J. 1988. Determination of comminution characteristics from single

particle breakage tests and its application to ball mill scale up. Trans Inst Min Metall, 97, C115 -

124.

SHI, F. N. & NAPIER-MUNN, T. J. 1996. A model for slurry rheology. International Journal of Mineral

Processing, 47, 103-123.

SHI, F. N. & NAPIER-MUNN, T. J. 1999. Estimation of shear rates inside a ball mill. International

Journal of Mineral Processing, 57, 167-183.

SHI, F. N. & NAPIER-MUNN, T. J. 2002. Effects of slurry rheology on industrial grinding performance.

International Journal of Mineral Processing, 65, 125-140.

SIDDALL, B., HENDERSON, G. & PUTLAND, B. Factors influencing sizing of SAG mills from drill

core samples. In: SME, ed. SAG Conference, 1996 Vancouver, BC.

STARK, S., PERKINS, T. & NAPIER-MUNN, T. J. JK drop weight parameters – a statistical analysis of

their accuracy and precision, and the effect on SAG mill comminution circuit simulation. In:

METALLURGY, T. A. I. O. M. A., ed. MetPlant 2008, 2008 Perth. 147 - 156.

STARKEY, J. & DOBBY, G. Application of the SAG Power Index at five Canadian SAG Plants. In: SME,

ed. Autogenous and Semi-Autogenous grinding conference, 1996 Vancouver, BC.

STARKEY, J., HINDSTROM, S. & NADSDY, G. SAG Design Testing, What it is and why it works. In:

SME, ed. SAG Conference 2006, 2006 Vancouver, BC.

ULUSAY, R. & HUDSON, J. A. 2007. The Complete ISRM Suggested Methods for Rock

Characterisation, Testing and Monitoring: 1974 - 2006, Ankara, Turkey, ISRM Turkish National

Group.

WIRFIYATAL, F. & MCCAFFERY, K. Applied Geo-Metallurgical Characterisation for Lif of Mine

Throughput Prediction at Batu Hijau. Proceedings of the Fifth International Conference on

Autogenous and Semiautogenous Grinding Technology, SAG 2011, 2011 Vancouver, Canada.